ssd.pytorch
ssd.pytorch copied to clipboard
amdegroot
A PyTorch Implementation of Single Shot MultiBox Detector
Hi, I don't really know about sh script, but when I try to download COCO dataset using COCO2014.sh, I found out that there's a missing space that cause the script...
Although issue #20 has said that this problem is solved in the latest version, I still have this problem, i.e., the number of Prior boxes are copied when it run...
how to solve this problem？ setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (6,) + inhomogeneous part.
**I convert the .pth model to .onnx model by this code:** ```python import torch import torch.onnx batch_size = 1 input_shape = (3, 300, 300) def convert_onnx(model): model.eval() dummy_input = torch.randn(batch_size,...
In augmetation file, at function `RandomSampleCrop`, we use np.random.choice, but I guess that there was some change in numpy lib or python so we can't use it for tupe like...
I want to get the **Precision** and **Recall** values after training, but eval.py can only get the **AP** and **mAP**. Does anyone have a solution to this problem?
In line 61 `flt[(rank < self.top_k).unsqueeze(-1).expand_as(flt)].fill_(0)` why `rank < self.top_k`? I think it should be `rank > self.top_k`. The position not in top_k need to be filled with value 0....
 I just adjust num_classes in config.py and ssd.py,why is still not correct? .how to solve this problem
# Problem In [match](https://github.com/amdegroot/ssd.pytorch/blob/8dd38657a3b1df98df26cf18be9671647905c2a0/layers/box_utils.py#L71) function, **which try to ensure every groundtruth boxe can be assigned to a prior boxe**(following code). But there is a problem , if there are same...
Hello great work there!! Can you help in adding negative samples while training. Images without bounding boxes or with bounding box and label as background
